My Issue: Why do we see what AI agents spent only after the allowance is gone?
My Idea: Botvisor™. Like SQL’s EXPLAIN feature, it would show the plan, estimated usage, uncertainty, and alternatives before the actual execution of the task/prompt.
It’s not a promise of savings, but instead it’s a chance to choose wisely before spending.
Read article below in my self-reply. Thanks!
Oh man, it's gotten worse, and it's gone from just being annoying to being frustrating. my usage reset today, and with just 2 tasks, 20% of my usage is gone already.
also, the bots have become a bit clumsy, work output is low-quality, no checks to verify results, and have to do multiple iterations even for simple tasks. it's really not getting any better at all. bit disappointing right now!🥹
IDEA: usage limits advisor, aka botvisor!
What if we had something similar to a sql-queryplan for our grokbots? We’d tell the bot what needs to be done, ask to mock-run it to estimate possible cost (percentage of your usage meter), and advise how to do it?
Also, remember the sci-fi movie “minority report”, and how the “precogs” were able to stop bad events before they happened? We gotta have something like that!
I mean, seriously, there’s a bunch of posts about how to make your bot-team more efficiently use their limits, but we all know that it’s still a challenge for most people and keep hitting limit.
So, gotta come up with something that will not only avoid cost of running bots being “unsustainably high relative to revenue”, but it’ll also be a great feature for users who understand the execution plan better, and use limits wisely. Huge win for everyone!
@elonmusk @grok your thoughts, please?
Berlin Engineering Center open — 3rd hub after Fremont & Shanghai. Tesla owner since 2020. Watching eng scale from my driveway.
#BREAKING: Regreta Corp (ticker: RGRT) plunges 68% in a single session after its flagship drug Regretamine is linked to “immediate and irreversible buyer’s remorse, and regret”, starting last week.
One desk head at a major bank was quoted saying: “We knew the name was a red flag. You don’t call a drug Regretamine and then act surprised when people regret taking it.”
The company, which had been trading near $142 after positive Phase II chatter on its ground-breaking compound Regretamine, collapsed in after-hours trading last night. By the open this morning the stock was already down more than 40%, and it finished the regular session at $45.20 on volume 11× the average.
More granular control over your bot team actions - love it!
Grok Bot now has a Permissions section in Settings!
It has two lists. One is Always Ask, for things your Grok Bot has to check with you first. The other is Allow without Asking, for things it can just do. Each list has its own Add Rule button. A Grok Bot can also offer a rule change on a card, and you confirm it.
A few rules you could write:
- Always ask before any outside e-mail goes out.
- Always ask before anything gets deleted.
- Always ask before it buys anything.
- Always ask before it overwrites any of your files.
- Allow reading your calendar without asking.
- Allow saving files on its own computer without asking.
- Allow simple read-only commands in one folder without asking.
If two rules conflict, the Always Ask rule wins, so it stops and asks you.
One more thing about buying. If your card is saved in the Vault, Grok Bot can fill it in without asking. So if you want to say yes to every payment yourself, don't save your card info there.
People could start handing Grok Bot bigger jobs now, because they can pick where Grok Bot has to stop and ask.
This is good for you because the small things you allow keep moving, and the things that matter still wait for your OK.
GrokBot just got a Permissions page. Gives you even more granular control over what you won't let your bot team do without your approval VS what they're allowed to do without asking.
My first default rule is ask approval for anything that you're going to touch like Mac-folders, GitHub-repos Vercel-projects, Neon-databases, or even your local cloned changelist. You can update later as needed!
Michigan winter is the real test, and Austin never asked FSD to find a lane under snow!
Detroit — the city where the automobile was born — is becoming ground zero for its next chapter.
The Detroit News reports today, in an exclusive, that Tesla is testing Cybercab robotaxis in Metro Detroit, joining Waymo, which already runs a robotaxi service there. No steering wheel, no pedals — the Cybercab is the first vehicle Musk has designed from day one for full autonomy, and it's now on the streets of America's automotive birthplace.
And this is just one piece of a remarkable November for $TSLA's robotaxi program:
- Cybercab expansion into Dallas is underway this month, per Tesla's Q3 earnings call
- Fort Worth went fully unsupervised on Nov 2 — the seventh driverless city, bringing the total network to 831 square miles
- Nevada regulators unanimously approved up to 5,000 fully autonomous vehicles in the Las Vegas area in year one
Why Detroit matters: it's where Ford and GM built the 20th century. If Tesla can put purpose-built robotaxis on those streets, it's a statement about where the 21st century is headed. And the data flywheel behind it keeps spinning — FSD v14 now runs on roughly 75% of North American deliveries, with more than 1.7 million active subscriptions feeding the system every single day.
Meanwhile, BMW is announcing urban driver assist "starting in 2027." Tesla's Cybercabs are already on Detroit streets.
This man is building the future — and right now, that future has a license plate in Michigan.













